lkml.org 
[lkml]   [2008]   [Sep]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC PATCH 1/3] Unified trace buffer

* Martin Bligh <mbligh@google.com> wrote:

> > And I don't think normalizing later is in any way more fundamentally
> > hard. It just means that you do part of the expensive things after
> > you have gathered the trace, rather than during.
>
> Agree with you on doing the expensive stuff later. If we wanted to get
> something that'd pack down to a couple fewer bits, and approximate ns,
> we could always >> 1 if you were > 2GHz, and >> 2 if you where > 4GHz,
> etc. which is at least cheap.

... which is exactly what sched_clock() does, combined with a
multiplication. (which is about as expensive as normal linear
arithmetics on most CPUs - i.e. in the 1 cycle range)

Normalizing has the advantage that we dont have to worry about it ever
again. Not about a changing scale due to cpufreq, slowing down or
speeding up TSCs due to C2/C3. We have so much TSC breakage all across
the spectrum that post-processing it is a nightmare in practice. Plus we
want sched_clock() to be fast anyway.

in the distant future we not only will have constant-TSC but it wont
stop in C2/C3 either at a whim (which they do right now, messing up
timestamps). At that stage fast time readout it will be so sane that CPU
makers should really provide a nanosec readout - it's easy to do a
simple multiplicator and hide the few cycles multiplicator latency to
RDTSC (this is continuous time after all so it's easy for the hw).

Hm?

Ingo


\
 
 \ /
  Last update: 2008-09-25 17:39    [from the cache]
©2003-2014 Jasper Spaans. Advertise on this site